【作者】 张洁；

【导师】 艾矫燕；

【作者基本信息】 广西大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 本文围绕植物黑腐病病斑，将计算机视觉与数字图像处理、模式识别技术结合起来，研究实现了黑腐病病斑的图像处理与检测。主要研究内容和研究结果如下：研究了国内外利用计算机视觉技术对农产品病害进行检测的研究进展和现状，并指出开展基于机器视觉的植物黑腐病病斑检测方法研究的优越性。建立并完善了适合本研究的图像采集的机器视觉系统，包括PC机，图像采集卡，工业COD彩色摄像头，扫描仪等。研究了图像噪声去除、图像增强、图像分割等多种图像处理算法，得到了适合叶片病斑快速分级的算法。对采集的病斑叶片图像进行做预处理。采用中值滤波消除图像噪声，此法简单易行，且具有较好的效果。采用小波变换来消除图像的不均匀照度问题。为去除背景对后续处理的影响，综合研究发现在Nrgb(Normalized RGB)模型的蓝色分量上，两峰间距离最大，选择b分量确定阈值进行背景分割。病斑图像分割方面，根据病斑图像特点，提出对H分量采用自适应模糊阈值算法来实现分割：采用自适应窗宽选择方法，利用动态窗宽选择阈值，结合两种已有的直方图变换技术，得出了一种新的直方图变换方法，在此基础上，利用模糊阈值分割法实现图像的有效分割。针对病斑特点，另外还采用了交互式手动分割方法。更多还原

【Abstract】 This study aims at the quick detection of leaf spots of crop black rot. Integrated the knowledge and technologies of computer vision, digital’image processing, and pattern recognition, the image processing and detection of leaf spots is come true. Main contents and results are shown as follows:The research advancements and achievements in the fields of farm produce disease using machine vision technique are reviewed. The superiority of leaf spots inspection method is put forward.A Machine vision system for leaf spots inspection has been set up. It included: Person Computer, frame grabber, industrial CCD color camera, scanner, etc.Several algorithms of image processing technique are studied, including elimination of image noise, image segmentation, image enhancement, etc. The method of image processing techniques can meet the demand of quick detection of leaf spots of crop black rot.The defected image are acquired and pre-processed. Image noise is eliminated using median filtering. The method has briefness and facility, strong ability of eliminating noise. Elimination of illumination unevenness is based wavelet. In order to eliminate the influence of background, the images are studied. The distance is maximal between two-peak values in blue gray level histogram of the Nrgb color model. Images are segmented using the threshold of b gray level.In the aspect of images segmentation methods, a method of self-adaptive fuzzy threshold segmentation on Hue image was proposed according to the characters of disease image. A method about self-adaptive selection of window-width is given, which improves the correctness of threshold selection. The dissertation obtains a new histogram transformation method by combining two existing histogram transformation methods, which can improve efficiency of fuzzy threshold segmentation method. According to the character of leaf spots, interactive hand segmentation is also used.更多还原